US12080013B2 - Multi-view depth estimation leveraging offline structure-from-motion - Google Patents

Abstract

A method for estimating depth of a scene includes selecting an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor of a first agent. The method also includes identifying previously captured images of the scene. The method further includes selecting a set of images from the previously captured images based on each image of the set of images satisfying depth criteria. The method still further includes estimating the depth of the scene based on the selected image and the selected set of images.

Description

CROSS-REFERENCE TO RELATED APPLICATION

The present application claims the benefit of U.S. Provisional Patent Application No. 63/048,366, filed on Jul. 6, 2021, and titled “MULTI-VIEW DEPTH ESTIMATION LEVERAGING OFFLINE STRUCTURE-FROM-MOTION,” the disclosure of which is expressly incorporated by reference in its entirety.

FIELD OF THE DISCLOSURE

Certain aspects of the present disclosure generally relate to scene reconstruction systems.

BACKGROUND

Autonomous agents (e.g., vehicles, robots, etc.) rely on machine vision for constructing a three-dimensional (3D) representation of a surrounding environment. The 3D representation may be used for various tasks, such as localization and/or autonomous navigation. The 3D representation accuracy improves the autonomous agent's ability to perform various tasks. It may be desirable to improve an accuracy of a 3D representation of an environment, where the 3D representation is constructed from one or more images captured by the autonomous agent.

SUMMARY

In one aspect of the present disclosure, a method for estimating a depth of a scene includes selecting an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor of a first agent. The method further includes identifying previously captured images of the scene. The method still further includes selecting a set of images from the previously captured images based on each image of the set of images satisfying depth criteria. The method also includes estimating the depth of the scene based on the selected image and the selected set of images.

Another aspect of the present disclosure is directed to an apparatus for estimating a depth of a scene. The apparatus includes selecting an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor of a first agent. The apparatus further includes identifying previously captured images of the scene. The apparatus still further includes selecting a set of images from the previously captured images based on each image of the set of images satisfying depth criteria. The apparatus also includes estimating the depth of the scene based on the selected image and the selected set of images.

In another aspect of the present disclosure, a non-transitory computer-readable medium with non-transitory program code recorded thereon for estimating a depth of a scene. The program code is executed by a processor and includes program code to select an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor of a first agent. The program code further includes program code to identify previously captured images of the scene. The program code still further includes program code to select a set of images from the previously captured images based on each image of the set of images satisfying depth criteria. The program code also includes program code to estimate the depth of the scene based on the selected image and the selected set of images.

Another aspect of the present disclosure is directed to an apparatus for estimating a depth of a scene. The apparatus includes a processor; a memory coupled with the processor; and instructions stored in the memory and operable, when executed by the processor, to cause the apparatus to select an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor of a first agent. Execution of the instructions further cause the apparatus to identify previously captured images of the scene. Execution of the instructions also cause the apparatus to select a set of images from the previously captured images based on each image of the set of images satisfying depth criteria. Execution of the instructions still further cause the apparatus to estimate the depth of the scene based on the selected image and the selected set of images.

The foregoing has outlined rather broadly the features and technical advantages of examples according to the disclosure in order that the detailed description that follows may be better understood. Additional features and advantages will be described. The conception and specific examples disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure. Such equivalent constructions do not depart from the scope of the appended claims. Characteristics of the concepts disclosed, both their organization and method of operation, together with associated advantages will be better understood from the following description when considered in connection with the accompanying figures. Each of the figures is provided for the purposes of illustration and description, and not as a definition of the limits of the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

The features, nature, and advantages of the present disclosure will become more apparent from the detailed description set forth below when taken in conjunction with the drawings in which like reference characters identify correspondingly throughout.

FIG.1 illustrates an example of a vehicle in an environment according to aspects of the present disclosure.

FIG.2A is an example of a single image according to aspects of the present disclosure.

FIG.2B is an example of a depth map according to aspects of the present disclosure.

FIG.2C is an example of a reconstructed target image according to aspects of the present disclosure.

FIG.3A illustrates an example of a depth prediction system according to aspects of the present disclosure.

FIG.3B is a diagram illustrating an example of matching keypoints and in a query image and a target image, in accordance with aspects of the present disclosure.

FIG.4 is a diagram illustrating an example of a hardware implementation for a depth estimation system, in accordance with aspects of the present disclosure.

FIG.5 illustrates a flow diagram for a process, in accordance with aspects of the present disclosure.

DETAILED DESCRIPTION

The detailed description set forth below, in connection with the appended drawings, is intended as a description of various configurations and is not intended to represent the only configurations in which the concepts described herein may be practiced. The detailed description includes specific details for the purpose of providing a thorough understanding of the various concepts. It will be apparent to those skilled in the art, however, that these concepts may be practiced without these specific details. In some instances, well-known structures and components are shown in block diagram form in order to avoid obscuring such concepts.

An agent, such as an autonomous agent, may generate a 3D representation of an environment based on one or more images obtained from a sensor. The 3D representation may also be referred to as a 3D model, a 3D scene, or a 3D map. 3D representations may facilitate various tasks, such as scene understanding, motion planning, and/or obstacle avoidance. For example, the agent may autonomously navigate through an environment based on the 3D representation.

In some examples, the 3D representation may be generated based on one or more depth maps of an environment. Each depth map may be generated based on one or more images of the environment captured by a sensor of an agent, such as an autonomous vehicle. In some such examples, the images may be 2D images. In some examples, the depth may be estimated from recently captured images. For example, the depth may be estimated from the last k images. In such examples, some of the last k images may provide depth information, while other images of the last k images may lack depth information. In such examples, a system for generating the 3D representation may not discriminate last k images when determining depth. Therefore, an accuracy of the 3D representation may be reduced.

It may be desirable to improve an accuracy of a 3D representation to improve tasks performed by the agent, such as scene understanding, motion planning, and/or obstacle avoidance. Aspects of the present disclosure improve an accuracy of a 3D representation of a location by estimating depth from images captured by both an agent and by one or more different agents. In some examples, each agent may be a vehicle, such as an autonomous vehicle. In some implementations, the images captured by the one or more different agents may be captured at a time prior to a time when the agent is generating the 3D representation for the location. The previously captured images may be associated with a dense map generated from multiple images captured by the one or more different agents at the location.

Aspects of the present disclosure are not limited to an autonomous agent. Aspects of the present disclosure also contemplate an agent operating in a manual mode or a semi-autonomous mode. In the manual mode, a human driver manually operates (e.g., controls) the agent. In the autonomous mode, an agent control system operates the agent without human intervention. In the semi-autonomous mode, the human may operate the agent, and the agent control system may override or assist the human. For example, the agent control system may override the human to prevent a collision or to obey one or more traffic rules.

FIG.1 illustrates an example of an ego vehicle100 (e.g., ego agent) in anenvironment150 according to aspects of the present disclosure. As shown inFIG.1, theego vehicle100 is traveling on aroad110. A first vehicle104 (e.g., other agent) may be ahead of theego vehicle100, and asecond vehicle116 may be adjacent to theego vehicle100. In this example, theego vehicle100 may include a2D camera108, such as a 2D RGB camera, and asecond sensor106. Thesecond sensor106 may be another RGB camera or another type of sensor, such as RADAR and/or ultrasound. Additionally, or alternatively, theego vehicle100 may include one or more additional sensors. For example, the additional sensors may be side facing and/or rear facing sensors.

In one configuration, the2D camera108 captures a 2D image that includes objects in the 2D camera's108 field ofview114. Thesecond sensor106 may generate one or more output streams. The 2D image captured by the 2D camera includes a 2D image of thefirst vehicle104, as thefirst vehicle104 is in the 2D camera's108 field ofview114.

The information obtained from thesensors106,108 may be used to navigate theego vehicle100 along a route when theego vehicle100 is in an autonomous mode. Thesensors106,108 may be powered from electricity provided from the vehicle's100 battery (not shown). The battery may also power the vehicle's motor. The information obtained from thesensors106,108 may be used to generate a 3D representation of an environment.

Aspects of the present disclosure are directed to a system for constructing a 3D representation of a location from one or more images of the location captured by a sensor, such as a monocular camera. In some examples, the sensor may be integrated with an agent, such as an autonomous vehicle. In one configuration, offline structure-from-motion is leveraged for multi-view depth estimation.

As described, the 3D representation may be generated based on one or more depth maps of an environment. Each depth map may be generated based on one or more images of the environment captured by a sensor of an agent, such as an autonomous vehicle. In some such examples, the images may be 2D images.

FIG.2A illustrates an example of atarget image200 of ascene202 according to aspects of the present disclosure. Thetarget image200 may be captured by a monocular camera. The monocular camera may capture a forward-facing view of an agent (e.g., a vehicle). In one configuration, the monocular camera is integrated with the vehicle. For example, the monocular camera may be defined in a roof structure, windshield, grill, or other portion of the vehicle. The vehicle may have one or more cameras and/or other types of sensors. Thetarget image200 may also be referred to as a current image. Thetarget image200 captures a 2D representation of a scene.

FIG.2B illustrates an example of adepth map220 of thescene202 according to aspects of the present disclosure. Thedepth map220 may be estimated from thetarget image200 and one or more source images. The source images may be images captured at a previous time step in relation to thetarget image200. Thedepth map220 provides a depth of a scene. The depth may be represented as a color or other feature.

FIG.2C illustrates an example of a3D reconstruction240 of thescene202 according to aspects of the present disclosure. The 3D reconstruction may be generated from thedepth map220 as well as a pose of thetarget image200 and a source image. As shown inFIGS.2A and2C, the viewing angle of thescene202 in the3D reconstruction240, is different from the viewing angle of thescene202 in thetarget image200. Because the3D reconstruction240 is a 3D view of thescene202, the viewing angle may be changed as desired. The3D reconstruction240 may be used to control one or more actions of the agent.

As shown in the examples ofFIGS.2A and2B, adepth map220 may be generated based on atarget image200. In some examples, the depth of an image may be estimated (e.g., generated) from two or more different images. For example, the depth of an image may be estimated from the last k images. In such examples, the last k images may be captured under the same environmental conditions, such as the same lighting. Therefore, an accuracy of the 3D representation may be reduced if the last k images fail to accurately capture an environment.

It may be desirable to improve an accuracy of a 3D representation to improve tasks performed by the agent, such as scene understanding, motion planning, and/or obstacle avoidance. Aspects of the present disclosure improve an accuracy of a 3D representation of a location by estimating depth from images captured by both an agent and by one or more different agents.

In some implementations, the system for constructing the 3D representation may be a component of the agent. Additionally, or alternatively, the system for constructing the 3D representation may be a cloud device, such as a cloud-based server, or another type of device. As an example, the system may be a model (e.g., an artificial neural network) trained to generate the 3D representation. The model may be implemented via one or more specially designed processors for generating the 3D representation and/or non-transitory computer-readable medium. For ease of explanation, operations of the system may be explained in terms of operation of the agent (e.g., ego agent).

FIG.3A is a block diagram illustrating an example of adepth prediction system300, in accordance with aspects of the present disclosure. As shown inFIG.3A, thedepth prediction system300 of anagent350 receives a sequence offrames302. Theagent350 may be an example of theego vehicle100 described with reference toFIG.1. The sequence offrames302 may be frames from a video captured by a sensor of the agent350 (not shown inFIG.3A). In some examples, the frames may be captured as theagent350 moves through an environment. Thedepth prediction system300 may be a non-generic processor that is specified for depth prediction.

In the example ofFIG.3A, aframe selection module320 of thedepth prediction system300 may select aframe304 from the sequence offrames302. In some examples, the selectedframe304 may be a keyframe. In some such examples, a depth map may be generated for the keyframe. As described, the depth map may be generated from two or more different images. In some implementations, thedepth prediction system300 transmits the selectedframe304 and/or a description of the selectedframe304 to aremote server352. The transmission may be performed over a wireless communication channel, such as a WiFi or cellular channel. As an example, the description may include keypoint descriptors of the selectedframe304. As another example, the description of the selectedframe304 may include location information, such as location coordinates associated with the selectedframe304, and pose information of theagent350 associated with the selectedframe304. The pose information may indicate a pose of theagent350 when the selectedframe304 was captured.

In some implementations, theremote server352 transmits a set of previously captured frames308 to theagent350 based on the selectedframe304 and/or a description of the selectedframe304. In some examples, one or more attributes of the set of previously captured frames308 match one or more attributes of the selectedframe304 and/or a description of the selectedframe304. In some such implementations, the one or more attributes may include key points, location information, or other attributes. As an example, key points of the selectedframe304 may match key points of the set of previously captured frames308. The set of previously captured frames308 may be selected from a number of previously captured frames stored at theremote server352.

In some implementations, theremote server352 generates a dense map based on the previously captured frames stored at theremote server352. In some examples, the dense map may be generated during an offline process. The dense map may be an example of a pre-generated map. Each frame of the previously captured frames may be captured at different times, or at a same time, from one or more agents.

In the example ofFIG.3A, afiltering module322 of thedepth prediction system300 may filter the set of previously captured frames308 to identify one or more frames310 that best describe an environment for the given task. The environment refers to an environment (e.g., spatial location) associated with the selectedframe304. In the current example, the given task is depth estimation, therefore, theagent350 may filter the set of previously captured frames308 to identify one or more frames310 that best describe an environment for depth estimation. That is, thefiltering module322 may filter the set of previously captured frames308 to identify one or more frames310 that satisfy depth criteria.

The set of previously captured frames308 may be filtered based on entropy (or other methods) to determine a subset of previously captured frames with depth information that satisfies depth criteria. Each frame may be an example of an image. In some examples, each frame satisfies the depth criteria based on the depth information of the frame being greater than a threshold. Additionally, or alternatively, each frame satisfies the depth criteria based on the frame having a greatest amount of depth information in comparison to other frames of the set of previously captured frames308.

In some implementations, thefiltering module322 may be a depth model trained to identify frames that satisfy depth criteria. The depth model may be an example of an artificial neural network. According to aspects of the present disclosure, filtering images to identify images that satisfy depth criteria may maximize an amount of obtainable depth information from a minimum number of images.

As shown in the example ofFIG.3A, the one or more frames310 that satisfy depth criteria may be provided to a depth estimation module356 to estimate the depth of the scene. In some examples, the depth estimation module356 estimates a depth of ascene312 based on the one or more frames310 and the selectedframe304. In some examples, the estimateddepth312 may be for a keyframe. The keyframe may be the selectedframe304 or one of the one or more frames310. The keyframe may be selected based on pre-defined triggers or learned triggers.

In some implementations, visual odometry methods maintain a window over a set of keyframes. A difference between each keyframe and adjacent frames in a sequence may be greater than a difference threshold. New keyframes may be added to the set of keyframes based on pre-defined triggers or learned triggers. For example, a keyframe may be added when theagent350 has satisfied motion criteria, such as moving X distance or rotating Y degrees.

As described, the depth estimation module356 receives a collection of frames I including the one or more frames310 and the selectedframe304, where I={I₁, I₂, . . . , I_N}. The depth estimation module356 may also receive the respective poses G corresponding to each frame I_tiof the collection of frame I, where G={G₁, G₂, . . . , G_N}. The depth estimation module356 predicts a dense depth map D*for the keyframe. The depth estimation module356 may be an example of an artificial neural network trained to estimate a depth based on two or more frames.

In some examples, the depth estimation module356 builds a cost volume over learned features. In such examples, information may be aggregated over multiple viewpoints by applying a global pooling layer that pools across viewpoints. Specifically, the depth module may be viewed as the composition of a 2D feature extractor, cost volume back-projection, and 3D stereo matching.

In some implementations, the depth estimation module356 begins by extracting learned features from the collection of frames I. The 2D encoder may include two stacked hourglass modules that map each image to a dense feature map. The cost volume back-projection selects a keyframe, such as frame I₁and a cost volume is constructed for each of the remaining N−1 frames. Additionally, the 3D stereo matching processes the set of N−1 cost volumes by a series of 3D convolutional layers to perform stereo matching. View pooling may then be performed by averaging over the N−1 cost volumes to aggregate information across frames. A single volume (e.g., aggregated volume) of dimensions H×W×D×2C may be generated based on the view pooling. The aggregated volume may be then processed by a series of 3D hourglass modules, each 3D hourglass module outputs an intermediate depth estimate.

The intermediate depth estimate may be generated by applying a 1×1×1 convolution to produce an H×W×D volume. A softmax operator may be applied over the depth dimension, so that a probability distribution over depths may be obtained for each pixel. The probability volume may be mapped into a single depth estimate using a differentiable argmax function which computes the expected depth.

The process for estimating the depth of the scene may be repeated for other frames captured at different times. In some examples, the selectedframe304 may include multiple frames. In some implementations, the depth model354 may be implemented at theremote server352. In such implementations, theremote server352 transmits the one or more frames310 that satisfy depth criteria to theagent350. In some other implementations, theremote server352 transmits multiple frames to theagent350 and theagent350 selects the set of previously captured frames308 from the multiple frames received from theremote server352.

In another implementation, the autonomous agent estimates the depth from a keyframe selected from the sequence offrames302. In one such implementation, the autonomous agent estimates the depth from the keyframe and one or more frames selected from the set of previously captured frames. In another such implementation, the autonomous agent estimates the depth based only on the selected keyframe.

Aspects of the present disclosure may improve the agent's350 invariance to changes in illumination and/or viewpoint. As described, in some examples, theagent350 may transmit a query, such as a query image, to theremote server352, such that theagent350 may estimate depth online (e.g., at the agent350). The selectedframe304 may be an example of the query image. In such examples, theremote server352 may perform a global localization search of a pre-generated map based on the query. The remote server may return a set of images308 that are similar to the query image. In some such examples, the agent may select one or more images310 from the set of images308 that have a maximum correlation to the query image. In such examples, theagent350 may estimate a depth of a current scene based on the selectedframe304 and the one or more selected images310.

As described, theremote server352 may perform the global localization search based on keypoints. In some examples, given a set of database images collected offline, learned keypoints and descriptors may be used to generate a globally consistent map. During online localization, theremote server352 may obtain the matching images from the database in response to a query image.FIG.3B is a diagram illustrating an example of matchingkeypoints366A and366B in aquery image360 and atarget image362, in accordance with aspects of the present disclosure. In the example ofFIG.3B, thequery image360 is an example of thequery frame304 described inFIG.3A. As described, a frame is an example of an image. Additionally, in the current example, thetarget image362 may be retrieved as a matching image associated with thequery image360 based on a number ofkeypoints366A in thequery image360 that match keypoints366B of thetarget image362. In some examples, thetarget image362 may be retrieved from a storage system, such as a memory device of the agent or a cloud-based storage system based on the number ofkeypoints366A in thequery image360. Thequery image360 may be unknown to the agent at a time when theimage360 is captured by a sensor of the agent.

In some examples, a distance between a location corresponding to thequery image360 and a location corresponding to thetarget image362 satisfies a distance condition. For example, the distance may satisfy the distance condition when the distance between the location of thequery image360 and the location of thetarget image362 is less than a distance threshold, such that theimages360 and362 are acquired in a same environment.

As described, in some implementations, thequery image360 may be used for a search an image database. The search may be an example of a visual search of the image database. In some examples, the images in the image database may be geotagged. Additionally, in some examples, the images of the database may include images of a sparse 3D model. That is, a sparse 3D model of an environment may be generated from two or more images stored in the image database.

In such implementations, locations of top-ranked images selected from the image database may be used as suggestions for a location of thequery image360. In some examples, an image extraction function produces a fixed size vector f(I_i) given a frame I_i, such as thequery image360. Specifically, the image extraction function may extract representations for the entire image database offline. At test time, the visual search is performed by finding a nearest database image to thequery image360 based on matching keypoints. In some examples, the image extraction function, or another module, may also extract features (e.g., keypoints) of thequery image360. The keypoints of thequery image360 may be matched with keypoints of one or more images in the image database. In one configuration, the keypoints may be matched on a global scale and then a local scale. Additionally, the keypoints may be directly matched. In some implementations, the nearest database image may be found via a nearest neighbor search, by sorting images based on a Euclidean distance between an image of the database and thequery image360. In the example ofFIG.3B, thetarget image362 may be the nearest database image to thequery image360.

In some implementations, a map of an agent's surrounding environment may be generated based on a depth estimate generated from thetarget image362 and thequery image360. In some examples, a six degree of freedom (6-DoF) pose of the sensor may be estimated from the matching keypoints. Additionally, or alternatively, in some examples, a sparse 3D representation, such as a 3D map, of the agent's surrounding environment may be generated based on the depth estimate. The pre-built map may be stored in a memory of the agent and/or stored in a remote memory location, such as a cloud memory storage location. The 3D maps may be used for localization and navigation.

As shown inFIG.3B, matchingkeypoints366A and366B between thequery image360 and thetarget image362 are depicted usingconnector lines364 between corresponding features.Connector lines364 may connect a center of receptive fields for matchingkeypoints366A and366B. Eachkeypoint366A and366B may correspond to a feature of animage360 and362. Thus, in some examples, a keypoint may be referred to as a feature. Additionally, as shown inFIG.3B, due to an angle of a sensor used for capturing an image, occlusion, and/or other factors, one or more keypoints366A and366B may not match. Additionally, in some implementations, eachkeypoint366A and366B may correspond with a respective descriptor. Each keypoint respective descriptor may be a string determined based on features of the image obtained from an encoding process of a keypoint model, such as a keypoint framework described with reference toFIG.3B.

In the example ofFIG.3B, theconnector lines364 are provided for illustrative purposes. A number ofconnector lines364 may correspond to a number of matching keypoints. In some examples, during testing (e.g., real-world deployment), theconnector lines364 may not be generated between matchedimages360 and362.

According to aspects of the present disclosure, a depth estimation system of the agent may be trained to estimate depth based on thequery image360 and a set of database images received from a remote server. The depth may be estimated based on multi-view stereo matching. In some implementations, offline map information may be leveraged to perform online depth estimation, thereby improving contextual information.

FIG.4 is a diagram illustrating an example of a hardware implementation for adepth estimation system400, according to aspects of the present disclosure. Thedepth estimation system400 may be a component of a vehicle, a robotic device, or another device. For example, as shown inFIG.4, thedepth estimation system400 is a component of avehicle428. Thevehicle428 may be an example of anego vehicle100 and anagent350 described with reference toFIGS.1 and3A, respectively. Aspects of the present disclosure are not limited to thedepth estimation system400 being a component of thevehicle428, as other types of agents, such as a bus, boat, drone, or robot, are also contemplated for using thedepth estimation system400.

Thevehicle428 may operate in one or more of an autonomous operating mode, a semi-autonomous operating mode, and a manual operating mode. Furthermore, thevehicle428 may be an electric vehicle, a hybrid vehicle, a fuel vehicle, or another type of vehicle.

Thedepth estimation system400 may be implemented with a bus architecture, represented generally by abus440. Thebus440 may include any number of interconnecting buses and bridges depending on the specific application of thedepth estimation system400 and the overall design constraints. Thebus440 links together various circuits including one or more processors and/or hardware modules, represented by aprocessor420, acommunication module422, alocation module418, asensor module402, alocomotion module426, anavigation module424, and a computer-readable medium414. Thebus440 may also link various other circuits such as timing sources, peripherals, voltage regulators, and power management circuits, which are well known in the art, and therefore, will not be described any further.

Thedepth estimation system400 includes atransceiver416 coupled to theprocessor420, thesensor module402, adepth estimation module408, thecommunication module422, thelocation module418, thelocomotion module426, thenavigation module424, and the computer-readable medium414. Thetransceiver416 is coupled to anantenna444. Thetransceiver416 communicates with various other devices over one or more communication networks, such as an infrastructure network, a V2V network, a V2I network, a V2X network, a V2P network, or another type of network. As an example, thetransceiver416 may transmit a warning to other agents when adepth estimation module408 determines that an occluded area is occluding high risk information.

Thedepth estimation system400 includes theprocessor420 coupled to the computer-readable medium414. Theprocessor420 performs processing, including the execution of software stored on the computer-readable medium414 providing functionality according to the disclosure. The software, when executed by theprocessor420, causes thedepth estimation system400 to perform the various functions described for a particular device, such as thevehicle428, or any of themodules402,408,414,414,418,420,422,424,424. The computer-readable medium414 may also be used for storing data that is manipulated by theprocessor420 when executing the software.

Thesensor module402 may be used to obtain measurements via different sensors, such as afirst sensor404 and asecond sensor406. Thefirst sensor404 may be a vision sensor, such as a stereoscopic camera or a red-green-blue (RGB) camera, for capturing 2D images. Thesecond sensor406 may be a ranging sensor, such as a light detection and ranging (LIDAR) sensor or a radio detection and ranging (RADAR) sensor. Of course, aspects of the present disclosure are not limited to the aforementioned sensors as other types of sensors, such as, for example, thermal, sonar, and/or lasers are also contemplated for either of thesensors404,404.

The measurements of thefirst sensor404 and thesecond sensor406 may be processed by one or more of theprocessor420, thesensor module402, thedepth estimation module408, thecommunication module422, thelocation module418, thelocomotion module426, thenavigation module424, in conjunction with the computer-readable medium414 to implement the functionality described herein. In one configuration, the data captured by thefirst sensor404 and thesecond sensor406 may be transmitted to an external device via thetransceiver416. Thefirst sensor404 and thesecond sensor406 may be coupled to thevehicle428 or may be in communication with thevehicle428.

Thelocation module418 may be used to determine a location of thevehicle428. For example, thelocation module418 may use a global positioning system (GPS) to determine the location of thevehicle428. Thecommunication module422 may be used to facilitate communications via thetransceiver416. For example, thecommunication module422 may be configured to provide communication capabilities via different wireless protocols, such as WiFi, long term evolution (LTE), 4G, etc. Thecommunication module422 may also be used to communicate with other components of thevehicle428 that are not modules of thedepth estimation system400.

Thelocomotion module426 may be used to facilitate locomotion of thevehicle428. As an example, thelocomotion module426 may control a movement of the wheels. As another example, thelocomotion module426 may be in communication with one or more power sources of thevehicle428, such as a motor and/or batteries. Of course, aspects of the present disclosure are not limited to providing locomotion via wheels and are contemplated for other types of components for providing locomotion, such as propellers, treads, fins, and/or jet engines.

Thedepth estimation system400 also includes thenavigation module424 for planning a route or controlling the locomotion of thevehicle428, via thelocomotion module426. Thenavigation module424 may override user input when the user input is expected (e.g., predicted) to cause a collision. The modules may be software modules running in theprocessor420, resident/stored in the computer-readable medium414, one or more hardware modules coupled to theprocessor420, or some combination thereof.

Thedepth estimation module408 may be in communication with and/or work in conjunction with one or more of thesensor module402, thetransceiver416, theprocessor420, thecommunication module422, thelocation module418, thelocomotion module426, thenavigation module424, and the computer-readable medium414. Thedepth estimation module408 may be configured to select an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor (e.g., thefirst sensor404 and/or the second sensor406) of thevehicle428. The selected image may comprises a keyframe. The image may be selected based on a pre-defined trigger or a learned trigger.

Thedepth estimation module408 may also be configured to identify a number of previously captured images of the scene. The images may be captured by one or more different agents different from thevehicle428. One or more of characteristics of the number of previously captured images are different from characteristics of the image of the scene selected from a sequence of images of the scene captured via an in-vehicle sensor of thevehicle428. The characteristics comprises at least pose and lighting. The previously captured images may be stored at a remote server.

Thedepth estimation module408 may further be configured to select a set of images from the number of images based on the set of images satisfying depth criteria. As an examples, the set of images may satisfy the depth criteria based on an amount of depth information for each image in the set of images is greater than a threshold. Additionally, or alternatively, the set of images may have a maximum correlation with the image of the scene. Thedepth estimation module408 may include a depth model trained to select the set of images.

Thedepth estimation module408 may additionally be configured to estimate the depth of the scene based on the selected image and the selected set of images. That is, thedepth estimation module408 may be trained for multi-view depth estimation. An action of thevehicle428 may be controlled based on the estimated depth.

FIG.5 illustrates a flow diagram for aprocess500 in accordance with aspects of the present disclosure. Theprocess500 may be executed by a vehicle, such as anego vehicle100, anagent350, and/or thedepth estimation module408, as described with reference toFIGS.1,3 and4, respectively. As shown inFIG.5, atblock502, theprocess500 selects an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor (e.g., thefirst sensor404 and/or the second sensor406) of a first agent. The selected image may comprises a keyframe. The image may be selected based on a pre-defined trigger or a learned trigger.

Theprocess500 may also identify a number of previously captured images of the scene (block504). The images may be captured by one or more different agents different from the first agent. One or more of characteristics of the number of previously captured images are different from characteristics of the image of the scene selected from a sequence of images of the scene captured via an in-vehicle sensor of the first agent. The characteristics comprises at least pose and lighting. The previously captured images may be stored at a remote server.

Theprocess500 may also select a set of images from the number of images based on the set of images satisfying depth criteria (block506). The set of images may be selected when the amount of depth information for each image in the set of images is greater than a threshold

Theprocess500 may further estimate the depth of the scene based on the selected image and the selected set of images (block508). An action of thevehicle428 may be controlled based on the estimated depth. As an example, thevehicle428 may be controlled to avoid an object, such as a pedestrian or an inanimate object on the road.

Based on the teachings, one skilled in the art should appreciate that the scope of the present disclosure is intended to cover any aspect of the present disclosure, whether implemented independently of or combined with any other aspect of the present disclosure. For example, an apparatus may be implemented or a method may be practiced using any number of the aspects set forth. In addition, the scope of the present disclosure is intended to cover such an apparatus or method practiced using other structure, functionality, or structure and functionality in addition to, or other than the various aspects of the present disclosure set forth. It should be understood that any aspect of the present disclosure may be embodied by one or more elements of a claim.

The word “exemplary” is used herein to mean “serving as an example, instance, or illustration.” Any aspect described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other aspects.

Although particular aspects are described herein, many variations and permutations of these aspects fall within the scope of the present disclosure. Although some benefits and advantages of the preferred aspects are mentioned, the scope of the present disclosure is not intended to be limited to particular benefits, uses or objectives. Rather, aspects of the present disclosure are intended to be broadly applicable to different technologies, system configurations, networks and protocols, some of which are illustrated by way of example in the figures and in the following description of the preferred aspects. The detailed description and drawings are merely illustrative of the present disclosure rather than limiting, the scope of the present disclosure being defined by the appended claims and equivalents thereof.

As used herein, the term “determining” encompasses a wide variety of actions. For example, “determining” may include calculating, computing, processing, deriving, investigating, looking up (e.g., looking up in a table, a database or another data structure), ascertaining and the like. Additionally, “determining” may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory) and the like. Furthermore, “determining” may include resolving, selecting, choosing, establishing, and the like.

As used herein, a phrase referring to “at least one of” a list of items refers to any combination of those items, including single members. As an example, “at least one of: a, b, or c” is intended to cover: a, b, c, a-b, a-c, b-c, and a-b-c.

The various illustrative logical blocks, modules and circuits described in connection with the present disclosure may be implemented or performed with a processor specially configured to perform the functions discussed in the present disclosure. The processor may be a neural network processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array signal (FPGA) or other programmable logic device (PLD), discrete gate or transistor logic, discrete hardware components or any combination thereof designed to perform the functions described herein. Alternatively, the processing system may comprise one or more neuromorphic processors for implementing the neuron models and models of neural systems described herein. The processor may be a microprocessor, controller, microcontroller, or state machine specially configured as described herein. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or such other special configuration, as described herein.

The steps of a method or algorithm described in connection with the present disclosure may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in storage or machine readable medium, including random access memory (RAM), read only memory (ROM), flash memory, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, a hard disk, a removable disk, a CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. A software module may comprise a single instruction, or many instructions, and may be distributed over several different code segments, among different programs, and across multiple storage media. A storage medium may be coupled to a processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor.

The methods disclosed herein comprise one or more steps or actions for achieving the described method. The method steps and/or actions may be interchanged with one another without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and/or use of specific steps and/or actions may be modified without departing from the scope of the claims.

The functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in hardware, an example hardware configuration may comprise a processing system in a device. The processing system may be implemented with a bus architecture. The bus may include any number of interconnecting buses and bridges depending on the specific application of the processing system and the overall design constraints. The bus may link together various circuits including a processor, machine-readable media, and a bus interface. The bus interface may be used to connect a network adapter, among other things, to the processing system via the bus. The network adapter may be used to implement signal processing functions. For certain aspects, a user interface (e.g., keypad, display, mouse, joystick, etc.) may also be connected to the bus. The bus may also link various other circuits such as timing sources, peripherals, voltage regulators, power management circuits, and the like, which are well known in the art, and therefore, will not be described any further.

The processor may be responsible for managing the bus and processing, including the execution of software stored on the machine-readable media. Software shall be construed to mean instructions, data, or any combination thereof, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

In a hardware implementation, the machine-readable media may be part of the processing system separate from the processor. However, as those skilled in the art will readily appreciate, the machine-readable media, or any portion thereof, may be external to the processing system. By way of example, the machine-readable media may include a transmission line, a carrier wave modulated by data, and/or a computer product separate from the device, all which may be accessed by the processor through the bus interface. Alternatively, or in addition, the machine-readable media, or any portion thereof, may be integrated into the processor, such as the case may be with cache and/or specialized register files. Although the various components discussed may be described as having a specific location, such as a local component, they may also be configured in various ways, such as certain components being configured as part of a distributed computing system.

The machine-readable media may comprise a number of software modules. The software modules may include a transmission module and a receiving module. Each software module may reside in a single storage device or be distributed across multiple storage devices. By way of example, a software module may be loaded into RAM from a hard drive when a triggering event occurs. During execution of the software module, the processor may load some of the instructions into cache to increase access speed. One or more cache lines may then be loaded into a special purpose register file for execution by the processor. When referring to the functionality of a software module below, it will be understood that such functionality is implemented by the processor when executing instructions from that software module. Furthermore, it should be appreciated that aspects of the present disclosure result in improvements to the functioning of the processor, computer, machine, or other system implementing such aspects.

If implemented in software, the functions may be stored or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media include both computer storage media and communication media including any storage medium that facilitates transfer of a computer program from one place to another.

Further, it should be appreciated that modules and/or other appropriate means for performing the methods and techniques described herein can be downloaded and/or otherwise obtained by a user terminal and/or base station as applicable. For example, such a device can be coupled to a server to facilitate the transfer of means for performing the methods described herein. Alternatively, various methods described herein can be provided via storage means, such that a user terminal and/or base station can obtain the various methods upon coupling or providing the storage means to the device. Moreover, any other suitable technique for providing the methods and techniques described herein to a device can be utilized.

It is to be understood that the claims are not limited to the precise configuration and components illustrated above. Various modifications, changes, and variations may be made in the arrangement, operation, and details of the methods and apparatus described above without departing from the scope of the claims.

Claims (20)

What is claimed is:

1. A method for estimating depth of a scene:

selecting an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor of a first agent;

identifying a plurality of previously captured images of the scene;

selecting a set of images from the plurality of previously captured images based on each image of the set of images satisfying depth criteria; and

estimating the depth of the scene based on the selected image and the selected set of images.

2. The method ofclaim 1, wherein the plurality of previously captured images are captured by one or more second agents different from the first agent.

3. The method ofclaim 1, wherein one or more of characteristics of the plurality of previously captured images are different from characteristics of the selected image of the scene.

4. The method ofclaim 3, wherein the characteristics comprises at least pose and lighting.

5. The method ofclaim 1, further comprising selecting the set of images when the amount of depth information for each image in the set of images is greater than a threshold.

6. The method ofclaim 1, wherein the plurality of previously captured images are stored at a remote server.

7. The method ofclaim 1, further comprising selecting the image based on a pre-defined trigger or a learned trigger.

8. The method ofclaim 1, wherein the selected image comprises a keyframe.

9. The method ofclaim 1, further comprising controlling an action of the first agent based on the estimated depth.

10. An apparatus for estimating depth of a scene:

a processor;

a memory coupled with the processor; and

instructions stored in the memory and operable, when executed by the processor, to cause the apparatus to:

select an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor of a first agent;

identify a plurality of previously captured images of the scene;

select a set of images from the plurality of previously captured images based on each image of the set of images satisfying depth criteria; and

estimate the depth of the scene based on the selected image and the selected set of images.

11. The apparatus ofclaim 10, wherein the plurality of previously captured images are captured by one or more second agents different from the first agent.

12. The apparatus ofclaim 10, wherein one or more of characteristics of the plurality of previously captured images are different from characteristics of the selected image of the scene.

13. The apparatus ofclaim 12, wherein the characteristics comprises at least pose and lighting.

14. The apparatus ofclaim 10, wherein execution of the instructions further cause the apparatus to select the set of images when the amount of depth information for each image in the set of images is greater than a threshold.

15. The apparatus ofclaim 10, wherein the plurality of previously captured images are stored at a remote server.

16. The apparatus ofclaim 10, wherein execution of the instructions further cause the apparatus to select the image based on a pre-defined trigger or a learned trigger.

17. The apparatus ofclaim 10, wherein the selected image comprises a keyframe.

18. The apparatus ofclaim 10, wherein execution of the instructions further cause the apparatus to control an action of the first agent based on the estimated depth.

19. A non-transitory computer-readable medium having program code recorded thereon for estimating depth of a scene, the program code executed by a processor and comprising

program code to select an image of the scene from a sequence of images of the scene captured via an in-vehicle sensor of a first agent;

program code to identify a plurality of previously captured images of the scene;

program code to select a set of images from the plurality of previously captured images based on each image of the set of images satisfying depth criteria; and

program code to estimate the depth of the scene based on the selected image and the selected set of images.

20. The non-transitory computer-readable medium ofclaim 19, wherein:

the plurality of previously captured images are captured by one or more second agents different from the first agent; and

one or more of characteristics of the plurality of previously captured images are different from characteristics of the selected image of the scene.

Images